TubeTK/Events/2010.10.19: Difference between revisions

From KitwarePublic
< TubeTK‎ | Events
Jump to navigationJump to search
No edit summary
 
(6 intermediate revisions by 2 users not shown)
Line 28: Line 28:


= Danielle =
= Danielle =
== Sliding organ registration ==
* Fixed "duh" problem from Thursday - extracting tangential and normal components before computing regularization
* Regularization smoothing in normal direction completed
* Added intensity distance function (mean squares)
Translated sphere - motion field following registration WITHOUT regularization:
[[File:translatedSphereMotionFieldWithoutRegularization.png| 1000px]]
Translated sphere - motion field following registration WITH regularization:
[[File:translatedSphereMotionFieldWithRegularization.png| 1000px]]
== XCAT phantom ==
* Got XCAT running on my machine (thanks Andinet!)
* Learned XCAT parameters
* Wrote isbiValidation.py:
** --basic (XCAT mode 0 - phantom mode): with respiratory period 5s and cardiac period 1s, generates two images at 0s and 2.5 s, saves as .mhd
** --gold (XCAT mode 2 - spherical lesion generation mode): creates gold standard motion field, and saves as .txt (index, vector in 6 columns).  Then run itkImageToImageDiffusiveDeformableRegistrationGenerateGoldStandardTest to create itk::Image<VectorType, Dimension> .mhd for visualization
*** XCAT phantom gives gold standard vectors for surface borders (XCAT mode 4), but no internal vectors!  So...
*** for index in range - generate spherical lesion at index, use nurbs_save = 1 to save lesion.nrb, then calculate centroid of lesion control points
[[File:2010_10_19_motionFieldParaviewOverview.png|800 px]]
*** Question: are there discontinuities / sliding motion.  Looks like it =)  See circled region:
[[File:2010_10_19_goldStandardMotionFieldDiscontinuities_annotated.png|800 px]]
*** Question: does the motion field make sense?  Yes:
Fixed image (look at origin of vectors):
[[File:2010_10_19_motionFieldParaview_1.png|800 px]]
Moving image (look at destination of vectors):
[[File:2010_10_19_motionFieldParaview_2.png|800 px]]
** --segmentation (XCAT mode 5 - save anatomical variation mode, or run XCAT mode 0 with save_nurbs = 1) to save surfaces (.nrb text file).  Then create new .txt file for the surface you want.  Then run itkImageToImageDiffusiveDeformableRegistrationGenerateNormalsTest to create itk::Image<PixelType, Dimension> of control points
*** Unfortunately: (1) control points need interpolation, can't find NURBS support in VTK; (2) XCAT has MANY types of surfaces for one organ - ex all intracardiac surfaces, etc, and it would be a lot of work to sort through them; (3) Perfect segmentation not great for registration method evaluation - removes segmentation error component of total error.
*** Ex: right lung
[[File:segmentation2_rlung.png | 800 px]]
== Talking points ==
* Underlying XCAT deformation fields too perfect?  They are likely interpolating the border vectors.
== Future ==
* Boundary conditions
* Compute weighting term = distance to organ border
* Extract organ border normals from input data
* Run registration + compare to gold standard for validation #s
[[Category:TubeTK Events and Meetings|2010.10.19]]

Latest revision as of 18:48, 26 July 2013

Stephen

Completed

  • Ridge extractor success
  • Email requests for TubeTK
  • Licensing changes
  • Blur3D -> Blur

Ongoing

  • Radius extractor tests
    • 2D and 3D only :(
  • Funding re-allocation

Upcoming

  • Application in plant biology
  • Proposal with Paul / Liz

Romain

  • Implementation on the ComputeTortuosity module (only Distance Metric)
    • Started debugging
  • Future
    • More debugging! ---> hopefully testing soon
    • #tube command
    • Implementation of the other 2 methods : ICM SOAM

Danielle

Sliding organ registration

  • Fixed "duh" problem from Thursday - extracting tangential and normal components before computing regularization
  • Regularization smoothing in normal direction completed
  • Added intensity distance function (mean squares)

Translated sphere - motion field following registration WITHOUT regularization:

TranslatedSphereMotionFieldWithoutRegularization.png

Translated sphere - motion field following registration WITH regularization:

TranslatedSphereMotionFieldWithRegularization.png

XCAT phantom

  • Got XCAT running on my machine (thanks Andinet!)
  • Learned XCAT parameters
  • Wrote isbiValidation.py:
    • --basic (XCAT mode 0 - phantom mode): with respiratory period 5s and cardiac period 1s, generates two images at 0s and 2.5 s, saves as .mhd
    • --gold (XCAT mode 2 - spherical lesion generation mode): creates gold standard motion field, and saves as .txt (index, vector in 6 columns). Then run itkImageToImageDiffusiveDeformableRegistrationGenerateGoldStandardTest to create itk::Image<VectorType, Dimension> .mhd for visualization
      • XCAT phantom gives gold standard vectors for surface borders (XCAT mode 4), but no internal vectors! So...
      • for index in range - generate spherical lesion at index, use nurbs_save = 1 to save lesion.nrb, then calculate centroid of lesion control points

2010 10 19 motionFieldParaviewOverview.png

      • Question: are there discontinuities / sliding motion. Looks like it =) See circled region:

2010 10 19 goldStandardMotionFieldDiscontinuities annotated.png

      • Question: does the motion field make sense? Yes:

Fixed image (look at origin of vectors): 2010 10 19 motionFieldParaview 1.png

Moving image (look at destination of vectors): 2010 10 19 motionFieldParaview 2.png

    • --segmentation (XCAT mode 5 - save anatomical variation mode, or run XCAT mode 0 with save_nurbs = 1) to save surfaces (.nrb text file). Then create new .txt file for the surface you want. Then run itkImageToImageDiffusiveDeformableRegistrationGenerateNormalsTest to create itk::Image<PixelType, Dimension> of control points
      • Unfortunately: (1) control points need interpolation, can't find NURBS support in VTK; (2) XCAT has MANY types of surfaces for one organ - ex all intracardiac surfaces, etc, and it would be a lot of work to sort through them; (3) Perfect segmentation not great for registration method evaluation - removes segmentation error component of total error.
      • Ex: right lung

Segmentation2 rlung.png

Talking points

  • Underlying XCAT deformation fields too perfect? They are likely interpolating the border vectors.

Future

  • Boundary conditions
  • Compute weighting term = distance to organ border
  • Extract organ border normals from input data
  • Run registration + compare to gold standard for validation #s